Patent attributes
A method of training a heterogeneous sensory system, including: accepting data of an environment from a first sensor, identifying objects in the data from the first sensor by a first processing unit, accepting data of the environment from a second sensor corresponding to the data of the first sensor, identifying objects in the data from the second sensor by a second processing unit, comparing the identified objects from the first processing unit and the corresponding objects from the second processing unit with a system processing unit, determining cases of discrepancies, between the objects identified in the data of the first sensor and not in the data of the second sensor or vice versa, tagging the data of the cases with discrepancies, storing the tagged data in a training buffer; and training the first processing unit and/or the second processing unit with the data stored in the training buffer.
